Implementing Position-Based Real-Time Simulation of Large Crowds

Various methods have been proposed for simulating crowds of agents in recent years. Regrettably, not all are computational scalable as the number of simulated agents grows. Such quality is particularly important for virtual production, gaming, and immersive reality platforms. In this work, we provide an open-source implementation for the recently proposed Position-based dynamics approach to crowd simulation. Position-based crowd simulation was proven to be real-time, and scalable for crowds of up to 100k agents, while retaining dynamic agent and group behaviors. We provide both non-parallel, and GPU-based implementations. Our implementation is demonstrated on several scenarios, including examples from the original work. We witness interactive computation run-times, as well as visually realistic collective behavior.